The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill exhibits a significant attack surface by ingesting untrusted simulation data which is then used to parameterize shell commands.
Ingestion points: Simulation output files such as results/field_0100.json and results/history.json (SKILL.md).
Boundary markers: Absent; no delimiters or instructions to ignore embedded content are present when processing external file content.
Capability inventory: Bash, Write, Read, and Python subprocess execution for analysis scripts (SKILL.md).
Sanitization: Absent; there is no evidence of path validation or field name sanitization.
[Dynamic Execution] (MEDIUM): The statistical_analyzer.py and derived_quantities.py scripts accept complex logical strings (e.g., --region "x>0.3 and x<0.7"). This indicates a potential use of eval() or exec() for runtime expression parsing, which can lead to code execution if input data influences these strings.
[Command Execution] (MEDIUM): The workflow uses the Bash tool to construct and execute Python commands via string interpolation. Without strict escaping or the use of argument lists, this pattern is susceptible to shell injection through malicious file paths or field metadata.

post-processing